Fix post merge test issues (#707)

* add another job and remove post merge tests

* fix bug

* use matrix

Author: micmelesse

.github/workflows/amd_perf_kernel_Integration_tests.yml

@@ -4,9 +4,9 @@ on:
   workflow_dispatch:
   pull_request:
     branches: [main_perf]
-    merge_group:
-      branches: [main_perf]
-      types: [checks_requested]
+  merge_group:
+    branches: [main_perf]
+    types: [checks_requested]
 
 concurrency:
   group: ${{ github.ref }}
@@ -98,7 +98,9 @@ jobs:
     timeout-minutes: 90
     strategy:
       matrix:
-        runner: ${{fromJson(needs.Runner-Preparation-AMD.outputs.matrix-HIP)}}
+        runner: ${{ fromJson(needs.Runner-Preparation-AMD.outputs.matrix-HIP) }}
+        triton_version: [local, upstream]
+      fail-fast: false # disables failing the entire job when one matrix entry fails
     container:
       image: rocm/pytorch:latest
       options: --device=/dev/kfd --device=/dev/dri --security-opt seccomp=unconfined --group-add video --user root
@@ -117,13 +119,26 @@ jobs:
         run: |
           python3 -m pip install --upgrade pip
           python3 -m pip install lit matplotlib pandas
-      - name: Install Triton
+      - name: Install Triton (local or upstream)
+        run: |
+          if [ "${{ matrix.triton_version }}" = "local" ]; then
+            echo "Installing Local Triton..."
+            pip uninstall -y triton
+            cd python
+            pip install -v -e .
+            pip install numpy==1.26.4
+          else
+            echo "Installing Upstream Triton..."
+            pip uninstall -y triton
+            pip install matplotlib pandas pytest
+            git clone https://github.com/triton-lang/triton
+            cd triton
+            pip install --verbose -e python
+            cd ..
+          fi
+      - name: Show Triton version
         run: |
-          echo "PATH is '$PATH'"
-          pip uninstall -y triton
-          cd python
-          pip install -v -e .
-          pip install numpy==1.26.4
+          pip show triton
       - name: Run Perf Kernels Unit Tests
         run: |
           pytest -vvv ./python/perf-kernels/flash-attention.py